[BioC] normalization for custom chip

Hinnerk Boriss boriss at izbi.uni-leipzig.de
Fri Nov 12 08:07:56 CET 2004

Dear Shibing,

I do not recommend using house keeping genes for normalization. In several
experiments they turn out being differentially expressed. A better approach
would be to use a normalization method that searches for an invariant set of
genes in the sample. "VSN" and Li & Wong's "invariant set" do that. The
methods have limits though regarding the minimum proportion of not
differentially expressed genes. Below 30% things become difficult. Another
aspect you should be aware of is that a bias in the treatment effect, i.e.
treatment causes either mostly up- or down-regulation of genes, will distort
your normalization. VSN is most robust against this bias. 

Just an idea for you chip design: make a list of all genes that you think
could react to the planned treatment for 70-80% of your probe sets, then
take a random sample from all the remaining genes (of which you have no
prior evidence for differential expression) to design the remaining 20-30%
of the chip. This should get you a way out your normalization problem
typical for custom chips. In fact, you could restrict the invariant set
algorithm to search only in the random selection of genes. 


-----Original Message-----
From: bioconductor-bounces at stat.math.ethz.ch
[mailto:bioconductor-bounces at stat.math.ethz.ch] On Behalf Of Deng, Shibing
Sent: Thursday, November 11, 2004 10:08 PM
To: 'bioconductor at stat.math.ethz.ch'
Subject: [BioC] normalization for custom chip

We are designing a custom Affymetrix chip with about 1700 genes. By design,
a large number of genes on the chip will be differentially expressed between
our treatment and control samples. The assumption for quantile normalization
and other distribution-based normalization methods will not hold for these
chips. To normalize them, we plan to put some "house-keeping" genes or
"invariant" genes covering a wide range of intensities on the chip. We are
not sure how many house-keeping genes we should have to get a good
normalization? I will appreciate your input on this issue.


LEGAL NOTICE\ Unless expressly stated otherwise, this messag...{{dropped}}

Bioconductor mailing list
Bioconductor at stat.math.ethz.ch

More information about the Bioconductor mailing list